포스트에디팅 측정지표를 통한 기계번역 오류 유형화 연구 (Revisiting Machine Translation Error Typology through Human Post-editing)

서지정보

· 발행기관 : 한국외국어대학교 통번역연구소
· 수록지 정보 : 통번역학연구 / 22권 / 1호 / 1 ~ 25페이지
· 저자명 : 곽중철, 한승희

초록

Post-editing refers to correcting and editing machine translation results, which means its fundamentals involve revision and quality assessment. Assessing machine translation quality is a subjective process that relies on human judgment, which is inferred from complex criteria and parameters in professional quality metrics. Thus, a growing motivation started to be placed on the creation of machine translation error typology based upon theoretical reviews. This paper examines theoretical approaches on criteria of translation review, quality assessment and MT post-editing, and then discusses machine translation errors based on the criteria of form and meaning (language use). The paper also analyzes error types with high frequency in machine translation and compares them with those in human translation. The major error types observed in machine translation are accuracy (language norm) and clarity (transfer of meaning), while stylistic errors such as consistency and redundancy affects readability in human translation. In particular, morphological errors in orthography and syntactic restructuring have a negative effect on machine translation results, which is presumably incurred by the way how MT processes and retrieves language data.

영어초록

Post-editing refers to correcting and editing machine translation results, which means its fundamentals involve revision and quality assessment. Assessing machine translation quality is a subjective process that relies on human judgment, which is inferred from complex criteria and parameters in professional quality metrics. Thus, a growing motivation started to be placed on the creation of machine translation error typology based upon theoretical reviews. This paper examines theoretical approaches on criteria of translation review, quality assessment and MT post-editing, and then discusses machine translation errors based on the criteria of form and meaning (language use). The paper also analyzes error types with high frequency in machine translation and compares them with those in human translation. The major error types observed in machine translation are accuracy (language norm) and clarity (transfer of meaning), while stylistic errors such as consistency and redundancy affects readability in human translation. In particular, morphological errors in orthography and syntactic restructuring have a negative effect on machine translation results, which is presumably incurred by the way how MT processes and retrieves language data.

참고자료

· 없음